kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/mm.h>
   9 #include <linux/slab.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/module.h>
  13 #include <linux/capability.h>
  14 #include <linux/completion.h>
  15 #include <linux/personality.h>
  16 #include <linux/tty.h>
  17 #include <linux/namespace.h>
  18 #include <linux/key.h>
  19 #include <linux/security.h>
  20 #include <linux/cpu.h>
  21 #include <linux/acct.h>
  22 #include <linux/file.h>
  23 #include <linux/binfmts.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/profile.h>
  26 #include <linux/mount.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/mempolicy.h>
  29 #include <linux/cpuset.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/signal.h>
  32 #include <linux/cn_proc.h>
  33 #include <linux/mutex.h>
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/unistd.h>
  37 #include <asm/pgtable.h>
  38 #include <asm/mmu_context.h>
  39
  40 extern void sem_exit (void);
  41 extern struct task_struct *child_reaper;
  42
  43 int getrusage(struct task_struct *, int, struct rusage __user *);
  44
  45 static void exit_mm(struct task_struct * tsk);
  46
  47 static void __unhash_process(struct task_struct *p)
  48 {
  49         nr_threads--;
  50         detach_pid(p, PIDTYPE_PID);
  51         detach_pid(p, PIDTYPE_TGID);
  52         if (thread_group_leader(p)) {
  53                 detach_pid(p, PIDTYPE_PGID);
  54                 detach_pid(p, PIDTYPE_SID);
  55                 if (p->pid)
  56                         __get_cpu_var(process_counts)--;
  57         }
  58
  59         REMOVE_LINKS(p);
  60 }
  61
  62 void release_task(struct task_struct * p)
  63 {
  64         int zap_leader;
  65         task_t *leader;
  66         struct dentry *proc_dentry;
  67
  68 repeat:
  69         atomic_dec(&p->user->processes);
  70         spin_lock(&p->proc_lock);
  71         proc_dentry = proc_pid_unhash(p);
  72         write_lock_irq(&tasklist_lock);
  73         if (unlikely(p->ptrace))
  74                 __ptrace_unlink(p);
  75         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  76         __exit_signal(p);
  77         /*
  78          * Note that the fastpath in sys_times depends on __exit_signal having
  79          * updated the counters before a task is removed from the tasklist of
  80          * the process by __unhash_process.
  81          */
  82         __unhash_process(p);
  83
  84         /*
  85          * If we are the last non-leader member of the thread
  86          * group, and the leader is zombie, then notify the
  87          * group leader's parent process. (if it wants notification.)
  88          */
  89         zap_leader = 0;
  90         leader = p->group_leader;
  91         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  92                 BUG_ON(leader->exit_signal == -1);
  93                 do_notify_parent(leader, leader->exit_signal);
  94                 /*
  95                  * If we were the last child thread and the leader has
  96                  * exited already, and the leader's parent ignores SIGCHLD,
  97                  * then we are the one who should release the leader.
  98                  *
  99                  * do_notify_parent() will have marked it self-reaping in
 100                  * that case.
 101                  */
 102                 zap_leader = (leader->exit_signal == -1);
 103         }
 104
 105         sched_exit(p);
 106         write_unlock_irq(&tasklist_lock);
 107         spin_unlock(&p->proc_lock);
 108         proc_pid_flush(proc_dentry);
 109         release_thread(p);
 110         put_task_struct(p);
 111
 112         p = leader;
 113         if (unlikely(zap_leader))
 114                 goto repeat;
 115 }
 116
 117 /* we are using it only for SMP init */
 118
 119 void unhash_process(struct task_struct *p)
 120 {
 121         struct dentry *proc_dentry;
 122
 123         spin_lock(&p->proc_lock);
 124         proc_dentry = proc_pid_unhash(p);
 125         write_lock_irq(&tasklist_lock);
 126         __unhash_process(p);
 127         write_unlock_irq(&tasklist_lock);
 128         spin_unlock(&p->proc_lock);
 129         proc_pid_flush(proc_dentry);
 130 }
 131
 132 /*
 133  * This checks not only the pgrp, but falls back on the pid if no
 134  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 135  * without this...
 136  */
 137 int session_of_pgrp(int pgrp)
 138 {
 139         struct task_struct *p;
 140         int sid = -1;
 141
 142         read_lock(&tasklist_lock);
 143         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 144                 if (p->signal->session > 0) {
 145                         sid = p->signal->session;
 146                         goto out;
 147                 }
 148         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 149         p = find_task_by_pid(pgrp);
 150         if (p)
 151                 sid = p->signal->session;
 152 out:
 153         read_unlock(&tasklist_lock);
 154
 155         return sid;
 156 }
 157
 158 /*
 159  * Determine if a process group is "orphaned", according to the POSIX
 160  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 161  * by terminal-generated stop signals.  Newly orphaned process groups are
 162  * to receive a SIGHUP and a SIGCONT.
 163  *
 164  * "I ask you, have you ever known what it is to be an orphan?"
 165  */
 166 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 167 {
 168         struct task_struct *p;
 169         int ret = 1;
 170
 171         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 172                 if (p == ignored_task
 173                                 || p->exit_state
 174                                 || p->real_parent->pid == 1)
 175                         continue;
 176                 if (process_group(p->real_parent) != pgrp
 177                             && p->real_parent->signal->session == p->signal->session) {
 178                         ret = 0;
 179                         break;
 180                 }
 181         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 182         return ret;     /* (sighing) "Often!" */
 183 }
 184
 185 int is_orphaned_pgrp(int pgrp)
 186 {
 187         int retval;
 188
 189         read_lock(&tasklist_lock);
 190         retval = will_become_orphaned_pgrp(pgrp, NULL);
 191         read_unlock(&tasklist_lock);
 192
 193         return retval;
 194 }
 195
 196 static inline int has_stopped_jobs(int pgrp)
 197 {
 198         int retval = 0;
 199         struct task_struct *p;
 200
 201         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 202                 if (p->state != TASK_STOPPED)
 203                         continue;
 204
 205                 /* If p is stopped by a debugger on a signal that won't
 206                    stop it, then don't count p as stopped.  This isn't
 207                    perfect but it's a good approximation.  */
 208                 if (unlikely (p->ptrace)
 209                     && p->exit_code != SIGSTOP
 210                     && p->exit_code != SIGTSTP
 211                     && p->exit_code != SIGTTOU
 212                     && p->exit_code != SIGTTIN)
 213                         continue;
 214
 215                 retval = 1;
 216                 break;
 217         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 218         return retval;
 219 }
 220
 221 /**
 222  * reparent_to_init - Reparent the calling kernel thread to the init task.
 223  *
 224  * If a kernel thread is launched as a result of a system call, or if
 225  * it ever exits, it should generally reparent itself to init so that
 226  * it is correctly cleaned up on exit.
 227  *
 228  * The various task state such as scheduling policy and priority may have
 229  * been inherited from a user process, so we reset them to sane values here.
 230  *
 231  * NOTE that reparent_to_init() gives the caller full capabilities.
 232  */
 233 static inline void reparent_to_init(void)
 234 {
 235         write_lock_irq(&tasklist_lock);
 236
 237         ptrace_unlink(current);
 238         /* Reparent to init */
 239         REMOVE_LINKS(current);
 240         current->parent = child_reaper;
 241         current->real_parent = child_reaper;
 242         SET_LINKS(current);
 243
 244         /* Set the exit signal to SIGCHLD so we signal init on exit */
 245         current->exit_signal = SIGCHLD;
 246
 247         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
 248                 set_user_nice(current, 0);
 249         /* cpus_allowed? */
 250         /* rt_priority? */
 251         /* signals? */
 252         security_task_reparent_to_init(current);
 253         memcpy(current->signal->rlim, init_task.signal->rlim,
 254                sizeof(current->signal->rlim));
 255         atomic_inc(&(INIT_USER->__count));
 256         write_unlock_irq(&tasklist_lock);
 257         switch_uid(INIT_USER);
 258 }
 259
 260 void __set_special_pids(pid_t session, pid_t pgrp)
 261 {
 262         struct task_struct *curr = current->group_leader;
 263
 264         if (curr->signal->session != session) {
 265                 detach_pid(curr, PIDTYPE_SID);
 266                 curr->signal->session = session;
 267                 attach_pid(curr, PIDTYPE_SID, session);
 268         }
 269         if (process_group(curr) != pgrp) {
 270                 detach_pid(curr, PIDTYPE_PGID);
 271                 curr->signal->pgrp = pgrp;
 272                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 273         }
 274 }
 275
 276 void set_special_pids(pid_t session, pid_t pgrp)
 277 {
 278         write_lock_irq(&tasklist_lock);
 279         __set_special_pids(session, pgrp);
 280         write_unlock_irq(&tasklist_lock);
 281 }
 282
 283 /*
 284  * Let kernel threads use this to say that they
 285  * allow a certain signal (since daemonize() will
 286  * have disabled all of them by default).
 287  */
 288 int allow_signal(int sig)
 289 {
 290         if (!valid_signal(sig) || sig < 1)
 291                 return -EINVAL;
 292
 293         spin_lock_irq(&current->sighand->siglock);
 294         sigdelset(&current->blocked, sig);
 295         if (!current->mm) {
 296                 /* Kernel threads handle their own signals.
 297                    Let the signal code know it'll be handled, so
 298                    that they don't get converted to SIGKILL or
 299                    just silently dropped */
 300                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 301         }
 302         recalc_sigpending();
 303         spin_unlock_irq(&current->sighand->siglock);
 304         return 0;
 305 }
 306
 307 EXPORT_SYMBOL(allow_signal);
 308
 309 int disallow_signal(int sig)
 310 {
 311         if (!valid_signal(sig) || sig < 1)
 312                 return -EINVAL;
 313
 314         spin_lock_irq(&current->sighand->siglock);
 315         sigaddset(&current->blocked, sig);
 316         recalc_sigpending();
 317         spin_unlock_irq(&current->sighand->siglock);
 318         return 0;
 319 }
 320
 321 EXPORT_SYMBOL(disallow_signal);
 322
 323 /*
 324  *      Put all the gunge required to become a kernel thread without
 325  *      attached user resources in one place where it belongs.
 326  */
 327
 328 void daemonize(const char *name, ...)
 329 {
 330         va_list args;
 331         struct fs_struct *fs;
 332         sigset_t blocked;
 333
 334         va_start(args, name);
 335         vsnprintf(current->comm, sizeof(current->comm), name, args);
 336         va_end(args);
 337
 338         /*
 339          * If we were started as result of loading a module, close all of the
 340          * user space pages.  We don't need them, and if we didn't close them
 341          * they would be locked into memory.
 342          */
 343         exit_mm(current);
 344
 345         set_special_pids(1, 1);
 346         down(&tty_sem);
 347         current->signal->tty = NULL;
 348         up(&tty_sem);
 349
 350         /* Block and flush all signals */
 351         sigfillset(&blocked);
 352         sigprocmask(SIG_BLOCK, &blocked, NULL);
 353         flush_signals(current);
 354
 355         /* Become as one with the init task */
 356
 357         exit_fs(current);       /* current->fs->count--; */
 358         fs = init_task.fs;
 359         current->fs = fs;
 360         atomic_inc(&fs->count);
 361         exit_files(current);
 362         current->files = init_task.files;
 363         atomic_inc(&current->files->count);
 364
 365         reparent_to_init();
 366 }
 367
 368 EXPORT_SYMBOL(daemonize);
 369
 370 static inline void close_files(struct files_struct * files)
 371 {
 372         int i, j;
 373         struct fdtable *fdt;
 374
 375         j = 0;
 376
 377         /*
 378          * It is safe to dereference the fd table without RCU or
 379          * ->file_lock because this is the last reference to the
 380          * files structure.
 381          */
 382         fdt = files_fdtable(files);
 383         for (;;) {
 384                 unsigned long set;
 385                 i = j * __NFDBITS;
 386                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 387                         break;
 388                 set = fdt->open_fds->fds_bits[j++];
 389                 while (set) {
 390                         if (set & 1) {
 391                                 struct file * file = xchg(&fdt->fd[i], NULL);
 392                                 if (file)
 393                                         filp_close(file, files);
 394                         }
 395                         i++;
 396                         set >>= 1;
 397                 }
 398         }
 399 }
 400
 401 struct files_struct *get_files_struct(struct task_struct *task)
 402 {
 403         struct files_struct *files;
 404
 405         task_lock(task);
 406         files = task->files;
 407         if (files)
 408                 atomic_inc(&files->count);
 409         task_unlock(task);
 410
 411         return files;
 412 }
 413
 414 void fastcall put_files_struct(struct files_struct *files)
 415 {
 416         struct fdtable *fdt;
 417
 418         if (atomic_dec_and_test(&files->count)) {
 419                 close_files(files);
 420                 /*
 421                  * Free the fd and fdset arrays if we expanded them.
 422                  * If the fdtable was embedded, pass files for freeing
 423                  * at the end of the RCU grace period. Otherwise,
 424                  * you can free files immediately.
 425                  */
 426                 fdt = files_fdtable(files);
 427                 if (fdt == &files->fdtab)
 428                         fdt->free_files = files;
 429                 else
 430                         kmem_cache_free(files_cachep, files);
 431                 free_fdtable(fdt);
 432         }
 433 }
 434
 435 EXPORT_SYMBOL(put_files_struct);
 436
 437 static inline void __exit_files(struct task_struct *tsk)
 438 {
 439         struct files_struct * files = tsk->files;
 440
 441         if (files) {
 442                 task_lock(tsk);
 443                 tsk->files = NULL;
 444                 task_unlock(tsk);
 445                 put_files_struct(files);
 446         }
 447 }
 448
 449 void exit_files(struct task_struct *tsk)
 450 {
 451         __exit_files(tsk);
 452 }
 453
 454 static inline void __put_fs_struct(struct fs_struct *fs)
 455 {
 456         /* No need to hold fs->lock if we are killing it */
 457         if (atomic_dec_and_test(&fs->count)) {
 458                 dput(fs->root);
 459                 mntput(fs->rootmnt);
 460                 dput(fs->pwd);
 461                 mntput(fs->pwdmnt);
 462                 if (fs->altroot) {
 463                         dput(fs->altroot);
 464                         mntput(fs->altrootmnt);
 465                 }
 466                 kmem_cache_free(fs_cachep, fs);
 467         }
 468 }
 469
 470 void put_fs_struct(struct fs_struct *fs)
 471 {
 472         __put_fs_struct(fs);
 473 }
 474
 475 static inline void __exit_fs(struct task_struct *tsk)
 476 {
 477         struct fs_struct * fs = tsk->fs;
 478
 479         if (fs) {
 480                 task_lock(tsk);
 481                 tsk->fs = NULL;
 482                 task_unlock(tsk);
 483                 __put_fs_struct(fs);
 484         }
 485 }
 486
 487 void exit_fs(struct task_struct *tsk)
 488 {
 489         __exit_fs(tsk);
 490 }
 491
 492 EXPORT_SYMBOL_GPL(exit_fs);
 493
 494 /*
 495  * Turn us into a lazy TLB process if we
 496  * aren't already..
 497  */
 498 static void exit_mm(struct task_struct * tsk)
 499 {
 500         struct mm_struct *mm = tsk->mm;
 501
 502         mm_release(tsk, mm);
 503         if (!mm)
 504                 return;
 505         /*
 506          * Serialize with any possible pending coredump.
 507          * We must hold mmap_sem around checking core_waiters
 508          * and clearing tsk->mm.  The core-inducing thread
 509          * will increment core_waiters for each thread in the
 510          * group with ->mm != NULL.
 511          */
 512         down_read(&mm->mmap_sem);
 513         if (mm->core_waiters) {
 514                 up_read(&mm->mmap_sem);
 515                 down_write(&mm->mmap_sem);
 516                 if (!--mm->core_waiters)
 517                         complete(mm->core_startup_done);
 518                 up_write(&mm->mmap_sem);
 519
 520                 wait_for_completion(&mm->core_done);
 521                 down_read(&mm->mmap_sem);
 522         }
 523         atomic_inc(&mm->mm_count);
 524         if (mm != tsk->active_mm) BUG();
 525         /* more a memory barrier than a real lock */
 526         task_lock(tsk);
 527         tsk->mm = NULL;
 528         up_read(&mm->mmap_sem);
 529         enter_lazy_tlb(mm, current);
 530         task_unlock(tsk);
 531         mmput(mm);
 532 }
 533
 534 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 535 {
 536         /*
 537          * Make sure we're not reparenting to ourselves and that
 538          * the parent is not a zombie.
 539          */
 540         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
 541         p->real_parent = reaper;
 542 }
 543
 544 static inline void reparent_thread(task_t *p, task_t *father, int traced)
 545 {
 546         /* We don't want people slaying init.  */
 547         if (p->exit_signal != -1)
 548                 p->exit_signal = SIGCHLD;
 549
 550         if (p->pdeath_signal)
 551                 /* We already hold the tasklist_lock here.  */
 552                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 553
 554         /* Move the child from its dying parent to the new one.  */
 555         if (unlikely(traced)) {
 556                 /* Preserve ptrace links if someone else is tracing this child.  */
 557                 list_del_init(&p->ptrace_list);
 558                 if (p->parent != p->real_parent)
 559                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 560         } else {
 561                 /* If this child is being traced, then we're the one tracing it
 562                  * anyway, so let go of it.
 563                  */
 564                 p->ptrace = 0;
 565                 list_del_init(&p->sibling);
 566                 p->parent = p->real_parent;
 567                 list_add_tail(&p->sibling, &p->parent->children);
 568
 569                 /* If we'd notified the old parent about this child's death,
 570                  * also notify the new parent.
 571                  */
 572                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 573                     thread_group_empty(p))
 574                         do_notify_parent(p, p->exit_signal);
 575                 else if (p->state == TASK_TRACED) {
 576                         /*
 577                          * If it was at a trace stop, turn it into
 578                          * a normal stop since it's no longer being
 579                          * traced.
 580                          */
 581                         ptrace_untrace(p);
 582                 }
 583         }
 584
 585         /*
 586          * process group orphan check
 587          * Case ii: Our child is in a different pgrp
 588          * than we are, and it was the only connection
 589          * outside, so the child pgrp is now orphaned.
 590          */
 591         if ((process_group(p) != process_group(father)) &&
 592             (p->signal->session == father->signal->session)) {
 593                 int pgrp = process_group(p);
 594
 595                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 596                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 597                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 598                 }
 599         }
 600 }
 601
 602 /*
 603  * When we die, we re-parent all our children.
 604  * Try to give them to another thread in our thread
 605  * group, and if no such member exists, give it to
 606  * the global child reaper process (ie "init")
 607  */
 608 static inline void forget_original_parent(struct task_struct * father,
 609                                           struct list_head *to_release)
 610 {
 611         struct task_struct *p, *reaper = father;
 612         struct list_head *_p, *_n;
 613
 614         do {
 615                 reaper = next_thread(reaper);
 616                 if (reaper == father) {
 617                         reaper = child_reaper;
 618                         break;
 619                 }
 620         } while (reaper->exit_state);
 621
 622         /*
 623          * There are only two places where our children can be:
 624          *
 625          * - in our child list
 626          * - in our ptraced child list
 627          *
 628          * Search them and reparent children.
 629          */
 630         list_for_each_safe(_p, _n, &father->children) {
 631                 int ptrace;
 632                 p = list_entry(_p,struct task_struct,sibling);
 633
 634                 ptrace = p->ptrace;
 635
 636                 /* if father isn't the real parent, then ptrace must be enabled */
 637                 BUG_ON(father != p->real_parent && !ptrace);
 638
 639                 if (father == p->real_parent) {
 640                         /* reparent with a reaper, real father it's us */
 641                         choose_new_parent(p, reaper, child_reaper);
 642                         reparent_thread(p, father, 0);
 643                 } else {
 644                         /* reparent ptraced task to its real parent */
 645                         __ptrace_unlink (p);
 646                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 647                             thread_group_empty(p))
 648                                 do_notify_parent(p, p->exit_signal);
 649                 }
 650
 651                 /*
 652                  * if the ptraced child is a zombie with exit_signal == -1
 653                  * we must collect it before we exit, or it will remain
 654                  * zombie forever since we prevented it from self-reap itself
 655                  * while it was being traced by us, to be able to see it in wait4.
 656                  */
 657                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 658                         list_add(&p->ptrace_list, to_release);
 659         }
 660         list_for_each_safe(_p, _n, &father->ptrace_children) {
 661                 p = list_entry(_p,struct task_struct,ptrace_list);
 662                 choose_new_parent(p, reaper, child_reaper);
 663                 reparent_thread(p, father, 1);
 664         }
 665 }
 666
 667 /*
 668  * Send signals to all our closest relatives so that they know
 669  * to properly mourn us..
 670  */
 671 static void exit_notify(struct task_struct *tsk)
 672 {
 673         int state;
 674         struct task_struct *t;
 675         struct list_head ptrace_dead, *_p, *_n;
 676
 677         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 678             && !thread_group_empty(tsk)) {
 679                 /*
 680                  * This occurs when there was a race between our exit
 681                  * syscall and a group signal choosing us as the one to
 682                  * wake up.  It could be that we are the only thread
 683                  * alerted to check for pending signals, but another thread
 684                  * should be woken now to take the signal since we will not.
 685                  * Now we'll wake all the threads in the group just to make
 686                  * sure someone gets all the pending signals.
 687                  */
 688                 read_lock(&tasklist_lock);
 689                 spin_lock_irq(&tsk->sighand->siglock);
 690                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 691                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 692                                 recalc_sigpending_tsk(t);
 693                                 if (signal_pending(t))
 694                                         signal_wake_up(t, 0);
 695                         }
 696                 spin_unlock_irq(&tsk->sighand->siglock);
 697                 read_unlock(&tasklist_lock);
 698         }
 699
 700         write_lock_irq(&tasklist_lock);
 701
 702         /*
 703          * This does two things:
 704          *
 705          * A.  Make init inherit all the child processes
 706          * B.  Check to see if any process groups have become orphaned
 707          *      as a result of our exiting, and if they have any stopped
 708          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 709          */
 710
 711         INIT_LIST_HEAD(&ptrace_dead);
 712         forget_original_parent(tsk, &ptrace_dead);
 713         BUG_ON(!list_empty(&tsk->children));
 714         BUG_ON(!list_empty(&tsk->ptrace_children));
 715
 716         /*
 717          * Check to see if any process groups have become orphaned
 718          * as a result of our exiting, and if they have any stopped
 719          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 720          *
 721          * Case i: Our father is in a different pgrp than we are
 722          * and we were the only connection outside, so our pgrp
 723          * is about to become orphaned.
 724          */
 725
 726         t = tsk->real_parent;
 727
 728         if ((process_group(t) != process_group(tsk)) &&
 729             (t->signal->session == tsk->signal->session) &&
 730             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 731             has_stopped_jobs(process_group(tsk))) {
 732                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 733                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 734         }
 735
 736         /* Let father know we died
 737          *
 738          * Thread signals are configurable, but you aren't going to use
 739          * that to send signals to arbitary processes.
 740          * That stops right now.
 741          *
 742          * If the parent exec id doesn't match the exec id we saved
 743          * when we started then we know the parent has changed security
 744          * domain.
 745          *
 746          * If our self_exec id doesn't match our parent_exec_id then
 747          * we have changed execution domain as these two values started
 748          * the same after a fork.
 749          *
 750          */
 751
 752         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 753             ( tsk->parent_exec_id != t->self_exec_id  ||
 754               tsk->self_exec_id != tsk->parent_exec_id)
 755             && !capable(CAP_KILL))
 756                 tsk->exit_signal = SIGCHLD;
 757
 758
 759         /* If something other than our normal parent is ptracing us, then
 760          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 761          * only has special meaning to our real parent.
 762          */
 763         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 764                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 765                 do_notify_parent(tsk, signal);
 766         } else if (tsk->ptrace) {
 767                 do_notify_parent(tsk, SIGCHLD);
 768         }
 769
 770         state = EXIT_ZOMBIE;
 771         if (tsk->exit_signal == -1 &&
 772             (likely(tsk->ptrace == 0) ||
 773              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 774                 state = EXIT_DEAD;
 775         tsk->exit_state = state;
 776
 777         write_unlock_irq(&tasklist_lock);
 778
 779         list_for_each_safe(_p, _n, &ptrace_dead) {
 780                 list_del_init(_p);
 781                 t = list_entry(_p,struct task_struct,ptrace_list);
 782                 release_task(t);
 783         }
 784
 785         /* If the process is dead, release it - nobody will wait for it */
 786         if (state == EXIT_DEAD)
 787                 release_task(tsk);
 788 }
 789
 790 fastcall NORET_TYPE void do_exit(long code)
 791 {
 792         struct task_struct *tsk = current;
 793         int group_dead;
 794
 795         profile_task_exit(tsk);
 796
 797         WARN_ON(atomic_read(&tsk->fs_excl));
 798
 799         if (unlikely(in_interrupt()))
 800                 panic("Aiee, killing interrupt handler!");
 801         if (unlikely(!tsk->pid))
 802                 panic("Attempted to kill the idle task!");
 803         if (unlikely(tsk->pid == 1))
 804                 panic("Attempted to kill init!");
 805         if (tsk->io_context)
 806                 exit_io_context();
 807
 808         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 809                 current->ptrace_message = code;
 810                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 811         }
 812
 813         /*
 814          * We're taking recursive faults here in do_exit. Safest is to just
 815          * leave this task alone and wait for reboot.
 816          */
 817         if (unlikely(tsk->flags & PF_EXITING)) {
 818                 printk(KERN_ALERT
 819                         "Fixing recursive fault but reboot is needed!\n");
 820                 set_current_state(TASK_UNINTERRUPTIBLE);
 821                 schedule();
 822         }
 823
 824         tsk->flags |= PF_EXITING;
 825
 826         /*
 827          * Make sure we don't try to process any timer firings
 828          * while we are already exiting.
 829          */
 830         tsk->it_virt_expires = cputime_zero;
 831         tsk->it_prof_expires = cputime_zero;
 832         tsk->it_sched_expires = 0;
 833
 834         if (unlikely(in_atomic()))
 835                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 836                                 current->comm, current->pid,
 837                                 preempt_count());
 838
 839         acct_update_integrals(tsk);
 840         if (tsk->mm) {
 841                 update_hiwater_rss(tsk->mm);
 842                 update_hiwater_vm(tsk->mm);
 843         }
 844         group_dead = atomic_dec_and_test(&tsk->signal->live);
 845         if (group_dead) {
 846                 hrtimer_cancel(&tsk->signal->real_timer);
 847                 exit_itimers(tsk->signal);
 848                 acct_process(code);
 849         }
 850         exit_mm(tsk);
 851
 852         exit_sem(tsk);
 853         __exit_files(tsk);
 854         __exit_fs(tsk);
 855         exit_namespace(tsk);
 856         exit_thread();
 857         cpuset_exit(tsk);
 858         exit_keys(tsk);
 859
 860         if (group_dead && tsk->signal->leader)
 861                 disassociate_ctty(1);
 862
 863         module_put(task_thread_info(tsk)->exec_domain->module);
 864         if (tsk->binfmt)
 865                 module_put(tsk->binfmt->module);
 866
 867         tsk->exit_code = code;
 868         proc_exit_connector(tsk);
 869         exit_notify(tsk);
 870 #ifdef CONFIG_NUMA
 871         mpol_free(tsk->mempolicy);
 872         tsk->mempolicy = NULL;
 873 #endif
 874         /*
 875          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 876          */
 877         mutex_debug_check_no_locks_held(tsk);
 878
 879         /* PF_DEAD causes final put_task_struct after we schedule. */
 880         preempt_disable();
 881         BUG_ON(tsk->flags & PF_DEAD);
 882         tsk->flags |= PF_DEAD;
 883
 884         schedule();
 885         BUG();
 886         /* Avoid "noreturn function does return".  */
 887         for (;;) ;
 888 }
 889
 890 EXPORT_SYMBOL_GPL(do_exit);
 891
 892 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 893 {
 894         if (comp)
 895                 complete(comp);
 896
 897         do_exit(code);
 898 }
 899
 900 EXPORT_SYMBOL(complete_and_exit);
 901
 902 asmlinkage long sys_exit(int error_code)
 903 {
 904         do_exit((error_code&0xff)<<8);
 905 }
 906
 907 task_t fastcall *next_thread(const task_t *p)
 908 {
 909         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 910 }
 911
 912 EXPORT_SYMBOL(next_thread);
 913
 914 /*
 915  * Take down every thread in the group.  This is called by fatal signals
 916  * as well as by sys_exit_group (below).
 917  */
 918 NORET_TYPE void
 919 do_group_exit(int exit_code)
 920 {
 921         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 922
 923         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 924                 exit_code = current->signal->group_exit_code;
 925         else if (!thread_group_empty(current)) {
 926                 struct signal_struct *const sig = current->signal;
 927                 struct sighand_struct *const sighand = current->sighand;
 928                 read_lock(&tasklist_lock);
 929                 spin_lock_irq(&sighand->siglock);
 930                 if (sig->flags & SIGNAL_GROUP_EXIT)
 931                         /* Another thread got here before we took the lock.  */
 932                         exit_code = sig->group_exit_code;
 933                 else {
 934                         sig->group_exit_code = exit_code;
 935                         zap_other_threads(current);
 936                 }
 937                 spin_unlock_irq(&sighand->siglock);
 938                 read_unlock(&tasklist_lock);
 939         }
 940
 941         do_exit(exit_code);
 942         /* NOTREACHED */
 943 }
 944
 945 /*
 946  * this kills every thread in the thread group. Note that any externally
 947  * wait4()-ing process will get the correct exit code - even if this
 948  * thread is not the thread group leader.
 949  */
 950 asmlinkage void sys_exit_group(int error_code)
 951 {
 952         do_group_exit((error_code & 0xff) << 8);
 953 }
 954
 955 static int eligible_child(pid_t pid, int options, task_t *p)
 956 {
 957         if (pid > 0) {
 958                 if (p->pid != pid)
 959                         return 0;
 960         } else if (!pid) {
 961                 if (process_group(p) != process_group(current))
 962                         return 0;
 963         } else if (pid != -1) {
 964                 if (process_group(p) != -pid)
 965                         return 0;
 966         }
 967
 968         /*
 969          * Do not consider detached threads that are
 970          * not ptraced:
 971          */
 972         if (p->exit_signal == -1 && !p->ptrace)
 973                 return 0;
 974
 975         /* Wait for all children (clone and not) if __WALL is set;
 976          * otherwise, wait for clone children *only* if __WCLONE is
 977          * set; otherwise, wait for non-clone children *only*.  (Note:
 978          * A "clone" child here is one that reports to its parent
 979          * using a signal other than SIGCHLD.) */
 980         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 981             && !(options & __WALL))
 982                 return 0;
 983         /*
 984          * Do not consider thread group leaders that are
 985          * in a non-empty thread group:
 986          */
 987         if (current->tgid != p->tgid && delay_group_leader(p))
 988                 return 2;
 989
 990         if (security_task_wait(p))
 991                 return 0;
 992
 993         return 1;
 994 }
 995
 996 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 997                                int why, int status,
 998                                struct siginfo __user *infop,
 999                                struct rusage __user *rusagep)
1000 {
1001         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1002         put_task_struct(p);
1003         if (!retval)
1004                 retval = put_user(SIGCHLD, &infop->si_signo);
1005         if (!retval)
1006                 retval = put_user(0, &infop->si_errno);
1007         if (!retval)
1008                 retval = put_user((short)why, &infop->si_code);
1009         if (!retval)
1010                 retval = put_user(pid, &infop->si_pid);
1011         if (!retval)
1012                 retval = put_user(uid, &infop->si_uid);
1013         if (!retval)
1014                 retval = put_user(status, &infop->si_status);
1015         if (!retval)
1016                 retval = pid;
1017         return retval;
1018 }
1019
1020 /*
1021  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1022  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1023  * the lock and this task is uninteresting.  If we return nonzero, we have
1024  * released the lock and the system call should return.
1025  */
1026 static int wait_task_zombie(task_t *p, int noreap,
1027                             struct siginfo __user *infop,
1028                             int __user *stat_addr, struct rusage __user *ru)
1029 {
1030         unsigned long state;
1031         int retval;
1032         int status;
1033
1034         if (unlikely(noreap)) {
1035                 pid_t pid = p->pid;
1036                 uid_t uid = p->uid;
1037                 int exit_code = p->exit_code;
1038                 int why, status;
1039
1040                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1041                         return 0;
1042                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1043                         return 0;
1044                 get_task_struct(p);
1045                 read_unlock(&tasklist_lock);
1046                 if ((exit_code & 0x7f) == 0) {
1047                         why = CLD_EXITED;
1048                         status = exit_code >> 8;
1049                 } else {
1050                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1051                         status = exit_code & 0x7f;
1052                 }
1053                 return wait_noreap_copyout(p, pid, uid, why,
1054                                            status, infop, ru);
1055         }
1056
1057         /*
1058          * Try to move the task's state to DEAD
1059          * only one thread is allowed to do this:
1060          */
1061         state = xchg(&p->exit_state, EXIT_DEAD);
1062         if (state != EXIT_ZOMBIE) {
1063                 BUG_ON(state != EXIT_DEAD);
1064                 return 0;
1065         }
1066         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1067                 /*
1068                  * This can only happen in a race with a ptraced thread
1069                  * dying on another processor.
1070                  */
1071                 return 0;
1072         }
1073
1074         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1075                 struct signal_struct *psig;
1076                 struct signal_struct *sig;
1077
1078                 /*
1079                  * The resource counters for the group leader are in its
1080                  * own task_struct.  Those for dead threads in the group
1081                  * are in its signal_struct, as are those for the child
1082                  * processes it has previously reaped.  All these
1083                  * accumulate in the parent's signal_struct c* fields.
1084                  *
1085                  * We don't bother to take a lock here to protect these
1086                  * p->signal fields, because they are only touched by
1087                  * __exit_signal, which runs with tasklist_lock
1088                  * write-locked anyway, and so is excluded here.  We do
1089                  * need to protect the access to p->parent->signal fields,
1090                  * as other threads in the parent group can be right
1091                  * here reaping other children at the same time.
1092                  */
1093                 spin_lock_irq(&p->parent->sighand->siglock);
1094                 psig = p->parent->signal;
1095                 sig = p->signal;
1096                 psig->cutime =
1097                         cputime_add(psig->cutime,
1098                         cputime_add(p->utime,
1099                         cputime_add(sig->utime,
1100                                     sig->cutime)));
1101                 psig->cstime =
1102                         cputime_add(psig->cstime,
1103                         cputime_add(p->stime,
1104                         cputime_add(sig->stime,
1105                                     sig->cstime)));
1106                 psig->cmin_flt +=
1107                         p->min_flt + sig->min_flt + sig->cmin_flt;
1108                 psig->cmaj_flt +=
1109                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1110                 psig->cnvcsw +=
1111                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1112                 psig->cnivcsw +=
1113                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1114                 spin_unlock_irq(&p->parent->sighand->siglock);
1115         }
1116
1117         /*
1118          * Now we are sure this task is interesting, and no other
1119          * thread can reap it because we set its state to EXIT_DEAD.
1120          */
1121         read_unlock(&tasklist_lock);
1122
1123         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1124         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1125                 ? p->signal->group_exit_code : p->exit_code;
1126         if (!retval && stat_addr)
1127                 retval = put_user(status, stat_addr);
1128         if (!retval && infop)
1129                 retval = put_user(SIGCHLD, &infop->si_signo);
1130         if (!retval && infop)
1131                 retval = put_user(0, &infop->si_errno);
1132         if (!retval && infop) {
1133                 int why;
1134
1135                 if ((status & 0x7f) == 0) {
1136                         why = CLD_EXITED;
1137                         status >>= 8;
1138                 } else {
1139                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1140                         status &= 0x7f;
1141                 }
1142                 retval = put_user((short)why, &infop->si_code);
1143                 if (!retval)
1144                         retval = put_user(status, &infop->si_status);
1145         }
1146         if (!retval && infop)
1147                 retval = put_user(p->pid, &infop->si_pid);
1148         if (!retval && infop)
1149                 retval = put_user(p->uid, &infop->si_uid);
1150         if (retval) {
1151                 // TODO: is this safe?
1152                 p->exit_state = EXIT_ZOMBIE;
1153                 return retval;
1154         }
1155         retval = p->pid;
1156         if (p->real_parent != p->parent) {
1157                 write_lock_irq(&tasklist_lock);
1158                 /* Double-check with lock held.  */
1159                 if (p->real_parent != p->parent) {
1160                         __ptrace_unlink(p);
1161                         // TODO: is this safe?
1162                         p->exit_state = EXIT_ZOMBIE;
1163                         /*
1164                          * If this is not a detached task, notify the parent.
1165                          * If it's still not detached after that, don't release
1166                          * it now.
1167                          */
1168                         if (p->exit_signal != -1) {
1169                                 do_notify_parent(p, p->exit_signal);
1170                                 if (p->exit_signal != -1)
1171                                         p = NULL;
1172                         }
1173                 }
1174                 write_unlock_irq(&tasklist_lock);
1175         }
1176         if (p != NULL)
1177                 release_task(p);
1178         BUG_ON(!retval);
1179         return retval;
1180 }
1181
1182 /*
1183  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1184  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1185  * the lock and this task is uninteresting.  If we return nonzero, we have
1186  * released the lock and the system call should return.
1187  */
1188 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1189                              struct siginfo __user *infop,
1190                              int __user *stat_addr, struct rusage __user *ru)
1191 {
1192         int retval, exit_code;
1193
1194         if (!p->exit_code)
1195                 return 0;
1196         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1197             p->signal && p->signal->group_stop_count > 0)
1198                 /*
1199                  * A group stop is in progress and this is the group leader.
1200                  * We won't report until all threads have stopped.
1201                  */
1202                 return 0;
1203
1204         /*
1205          * Now we are pretty sure this task is interesting.
1206          * Make sure it doesn't get reaped out from under us while we
1207          * give up the lock and then examine it below.  We don't want to
1208          * keep holding onto the tasklist_lock while we call getrusage and
1209          * possibly take page faults for user memory.
1210          */
1211         get_task_struct(p);
1212         read_unlock(&tasklist_lock);
1213
1214         if (unlikely(noreap)) {
1215                 pid_t pid = p->pid;
1216                 uid_t uid = p->uid;
1217                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1218
1219                 exit_code = p->exit_code;
1220                 if (unlikely(!exit_code) ||
1221                     unlikely(p->state & TASK_TRACED))
1222                         goto bail_ref;
1223                 return wait_noreap_copyout(p, pid, uid,
1224                                            why, (exit_code << 8) | 0x7f,
1225                                            infop, ru);
1226         }
1227
1228         write_lock_irq(&tasklist_lock);
1229
1230         /*
1231          * This uses xchg to be atomic with the thread resuming and setting
1232          * it.  It must also be done with the write lock held to prevent a
1233          * race with the EXIT_ZOMBIE case.
1234          */
1235         exit_code = xchg(&p->exit_code, 0);
1236         if (unlikely(p->exit_state)) {
1237                 /*
1238                  * The task resumed and then died.  Let the next iteration
1239                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1240                  * already be zero here if it resumed and did _exit(0).
1241                  * The task itself is dead and won't touch exit_code again;
1242                  * other processors in this function are locked out.
1243                  */
1244                 p->exit_code = exit_code;
1245                 exit_code = 0;
1246         }
1247         if (unlikely(exit_code == 0)) {
1248                 /*
1249                  * Another thread in this function got to it first, or it
1250                  * resumed, or it resumed and then died.
1251                  */
1252                 write_unlock_irq(&tasklist_lock);
1253 bail_ref:
1254                 put_task_struct(p);
1255                 /*
1256                  * We are returning to the wait loop without having successfully
1257                  * removed the process and having released the lock. We cannot
1258                  * continue, since the "p" task pointer is potentially stale.
1259                  *
1260                  * Return -EAGAIN, and do_wait() will restart the loop from the
1261                  * beginning. Do _not_ re-acquire the lock.
1262                  */
1263                 return -EAGAIN;
1264         }
1265
1266         /* move to end of parent's list to avoid starvation */
1267         remove_parent(p);
1268         add_parent(p, p->parent);
1269
1270         write_unlock_irq(&tasklist_lock);
1271
1272         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1273         if (!retval && stat_addr)
1274                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1275         if (!retval && infop)
1276                 retval = put_user(SIGCHLD, &infop->si_signo);
1277         if (!retval && infop)
1278                 retval = put_user(0, &infop->si_errno);
1279         if (!retval && infop)
1280                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1281                                           ? CLD_TRAPPED : CLD_STOPPED),
1282                                   &infop->si_code);
1283         if (!retval && infop)
1284                 retval = put_user(exit_code, &infop->si_status);
1285         if (!retval && infop)
1286                 retval = put_user(p->pid, &infop->si_pid);
1287         if (!retval && infop)
1288                 retval = put_user(p->uid, &infop->si_uid);
1289         if (!retval)
1290                 retval = p->pid;
1291         put_task_struct(p);
1292
1293         BUG_ON(!retval);
1294         return retval;
1295 }
1296
1297 /*
1298  * Handle do_wait work for one task in a live, non-stopped state.
1299  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1300  * the lock and this task is uninteresting.  If we return nonzero, we have
1301  * released the lock and the system call should return.
1302  */
1303 static int wait_task_continued(task_t *p, int noreap,
1304                                struct siginfo __user *infop,
1305                                int __user *stat_addr, struct rusage __user *ru)
1306 {
1307         int retval;
1308         pid_t pid;
1309         uid_t uid;
1310
1311         if (unlikely(!p->signal))
1312                 return 0;
1313
1314         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1315                 return 0;
1316
1317         spin_lock_irq(&p->sighand->siglock);
1318         /* Re-check with the lock held.  */
1319         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1320                 spin_unlock_irq(&p->sighand->siglock);
1321                 return 0;
1322         }
1323         if (!noreap)
1324                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1325         spin_unlock_irq(&p->sighand->siglock);
1326
1327         pid = p->pid;
1328         uid = p->uid;
1329         get_task_struct(p);
1330         read_unlock(&tasklist_lock);
1331
1332         if (!infop) {
1333                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1334                 put_task_struct(p);
1335                 if (!retval && stat_addr)
1336                         retval = put_user(0xffff, stat_addr);
1337                 if (!retval)
1338                         retval = p->pid;
1339         } else {
1340                 retval = wait_noreap_copyout(p, pid, uid,
1341                                              CLD_CONTINUED, SIGCONT,
1342                                              infop, ru);
1343                 BUG_ON(retval == 0);
1344         }
1345
1346         return retval;
1347 }
1348
1349
1350 static inline int my_ptrace_child(struct task_struct *p)
1351 {
1352         if (!(p->ptrace & PT_PTRACED))
1353                 return 0;
1354         if (!(p->ptrace & PT_ATTACHED))
1355                 return 1;
1356         /*
1357          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1358          * we are the attacher.  If we are the real parent, this is a race
1359          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1360          * which we have to switch the parent links, but has already set
1361          * the flags in p->ptrace.
1362          */
1363         return (p->parent != p->real_parent);
1364 }
1365
1366 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1367                     int __user *stat_addr, struct rusage __user *ru)
1368 {
1369         DECLARE_WAITQUEUE(wait, current);
1370         struct task_struct *tsk;
1371         int flag, retval;
1372
1373         add_wait_queue(&current->signal->wait_chldexit,&wait);
1374 repeat:
1375         /*
1376          * We will set this flag if we see any child that might later
1377          * match our criteria, even if we are not able to reap it yet.
1378          */
1379         flag = 0;
1380         current->state = TASK_INTERRUPTIBLE;
1381         read_lock(&tasklist_lock);
1382         tsk = current;
1383         do {
1384                 struct task_struct *p;
1385                 struct list_head *_p;
1386                 int ret;
1387
1388                 list_for_each(_p,&tsk->children) {
1389                         p = list_entry(_p,struct task_struct,sibling);
1390
1391                         ret = eligible_child(pid, options, p);
1392                         if (!ret)
1393                                 continue;
1394
1395                         switch (p->state) {
1396                         case TASK_TRACED:
1397                                 /*
1398                                  * When we hit the race with PTRACE_ATTACH,
1399                                  * we will not report this child.  But the
1400                                  * race means it has not yet been moved to
1401                                  * our ptrace_children list, so we need to
1402                                  * set the flag here to avoid a spurious ECHILD
1403                                  * when the race happens with the only child.
1404                                  */
1405                                 flag = 1;
1406                                 if (!my_ptrace_child(p))
1407                                         continue;
1408                                 /*FALLTHROUGH*/
1409                         case TASK_STOPPED:
1410                                 /*
1411                                  * It's stopped now, so it might later
1412                                  * continue, exit, or stop again.
1413                                  */
1414                                 flag = 1;
1415                                 if (!(options & WUNTRACED) &&
1416                                     !my_ptrace_child(p))
1417                                         continue;
1418                                 retval = wait_task_stopped(p, ret == 2,
1419                                                            (options & WNOWAIT),
1420                                                            infop,
1421                                                            stat_addr, ru);
1422                                 if (retval == -EAGAIN)
1423                                         goto repeat;
1424                                 if (retval != 0) /* He released the lock.  */
1425                                         goto end;
1426                                 break;
1427                         default:
1428                         // case EXIT_DEAD:
1429                                 if (p->exit_state == EXIT_DEAD)
1430                                         continue;
1431                         // case EXIT_ZOMBIE:
1432                                 if (p->exit_state == EXIT_ZOMBIE) {
1433                                         /*
1434                                          * Eligible but we cannot release
1435                                          * it yet:
1436                                          */
1437                                         if (ret == 2)
1438                                                 goto check_continued;
1439                                         if (!likely(options & WEXITED))
1440                                                 continue;
1441                                         retval = wait_task_zombie(
1442                                                 p, (options & WNOWAIT),
1443                                                 infop, stat_addr, ru);
1444                                         /* He released the lock.  */
1445                                         if (retval != 0)
1446                                                 goto end;
1447                                         break;
1448                                 }
1449 check_continued:
1450                                 /*
1451                                  * It's running now, so it might later
1452                                  * exit, stop, or stop and then continue.
1453                                  */
1454                                 flag = 1;
1455                                 if (!unlikely(options & WCONTINUED))
1456                                         continue;
1457                                 retval = wait_task_continued(
1458                                         p, (options & WNOWAIT),
1459                                         infop, stat_addr, ru);
1460                                 if (retval != 0) /* He released the lock.  */
1461                                         goto end;
1462                                 break;
1463                         }
1464                 }
1465                 if (!flag) {
1466                         list_for_each(_p, &tsk->ptrace_children) {
1467                                 p = list_entry(_p, struct task_struct,
1468                                                 ptrace_list);
1469                                 if (!eligible_child(pid, options, p))
1470                                         continue;
1471                                 flag = 1;
1472                                 break;
1473                         }
1474                 }
1475                 if (options & __WNOTHREAD)
1476                         break;
1477                 tsk = next_thread(tsk);
1478                 if (tsk->signal != current->signal)
1479                         BUG();
1480         } while (tsk != current);
1481
1482         read_unlock(&tasklist_lock);
1483         if (flag) {
1484                 retval = 0;
1485                 if (options & WNOHANG)
1486                         goto end;
1487                 retval = -ERESTARTSYS;
1488                 if (signal_pending(current))
1489                         goto end;
1490                 schedule();
1491                 goto repeat;
1492         }
1493         retval = -ECHILD;
1494 end:
1495         current->state = TASK_RUNNING;
1496         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1497         if (infop) {
1498                 if (retval > 0)
1499                 retval = 0;
1500                 else {
1501                         /*
1502                          * For a WNOHANG return, clear out all the fields
1503                          * we would set so the user can easily tell the
1504                          * difference.
1505                          */
1506                         if (!retval)
1507                                 retval = put_user(0, &infop->si_signo);
1508                         if (!retval)
1509                                 retval = put_user(0, &infop->si_errno);
1510                         if (!retval)
1511                                 retval = put_user(0, &infop->si_code);
1512                         if (!retval)
1513                                 retval = put_user(0, &infop->si_pid);
1514                         if (!retval)
1515                                 retval = put_user(0, &infop->si_uid);
1516                         if (!retval)
1517                                 retval = put_user(0, &infop->si_status);
1518                 }
1519         }
1520         return retval;
1521 }
1522
1523 asmlinkage long sys_waitid(int which, pid_t pid,
1524                            struct siginfo __user *infop, int options,
1525                            struct rusage __user *ru)
1526 {
1527         long ret;
1528
1529         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1530                 return -EINVAL;
1531         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1532                 return -EINVAL;
1533
1534         switch (which) {
1535         case P_ALL:
1536                 pid = -1;
1537                 break;
1538         case P_PID:
1539                 if (pid <= 0)
1540                         return -EINVAL;
1541                 break;
1542         case P_PGID:
1543                 if (pid <= 0)
1544                         return -EINVAL;
1545                 pid = -pid;
1546                 break;
1547         default:
1548                 return -EINVAL;
1549         }
1550
1551         ret = do_wait(pid, options, infop, NULL, ru);
1552
1553         /* avoid REGPARM breakage on x86: */
1554         prevent_tail_call(ret);
1555         return ret;
1556 }
1557
1558 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1559                           int options, struct rusage __user *ru)
1560 {
1561         long ret;
1562
1563         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1564                         __WNOTHREAD|__WCLONE|__WALL))
1565                 return -EINVAL;
1566         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1567
1568         /* avoid REGPARM breakage on x86: */
1569         prevent_tail_call(ret);
1570         return ret;
1571 }
1572
1573 #ifdef __ARCH_WANT_SYS_WAITPID
1574
1575 /*
1576  * sys_waitpid() remains for compatibility. waitpid() should be
1577  * implemented by calling sys_wait4() from libc.a.
1578  */
1579 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1580 {
1581         return sys_wait4(pid, stat_addr, options, NULL);
1582 }
1583
1584 #endif